4100 AWL/Thomas_ch14p965-1066 9/2/04 11:09 AM Page 1027

14.7 Extreme Values and Saddle Points 1027

14.7 Extreme Values and Saddle Points

z
Continuous functions of two variables assume extreme values on closed, bounded do-
mains (see Figures 14.36 and 14.37). We see in this section that we can narrow the search
for these extreme values by examining the functions’ first partial derivatives. A function of
two variables can assume extreme values only at domain boundary points or at interior do-
main points where both first partial derivatives are zero or where one or both of the first
partial derivatives fails to exist. However, the vanishing of derivatives at an interior point
(a, b) does not always signal the presence of an extreme value. The surface that is the
graph of the function might be shaped like a saddle right above (a, b) and cross its tangent
plane there.
x y

Derivative Tests for Local Extreme Values

FIGURE 14.36 The function To find the local extreme values of a function of a single variable, we look for points
z = scos xdscos yde -2x 2 + y 2 where the graph has a horizontal tangent line. At such points, we then look for local max-
ima, local minima, and points of inflection. For a function ƒ(x, y) of two variables, we look
has a maximum value of 1 and a minimum for points where the surface z = ƒsx, yd has a horizontal tangent plane. At such points, we
value of about -0.067 on the square then look for local maxima, local minima, and saddle points (more about saddle points in a
region ƒ x ƒ … 3p>2, ƒ y ƒ … 3p>2 . moment).

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 8/25/04 2:53 PM Page 1028

1028 Chapter 14: Partial Derivatives

z
DEFINITIONS Local Maximum, Local Minimum
Let ƒ(x, y) be defined on a region R containing the point (a, b). Then
1. ƒ(a, b) is a local maximum value of ƒ if ƒsa, bd Ú ƒsx, yd for all domain
points (x, y) in an open disk centered at (a, b).
2. ƒ(a, b) is a local minimum value of ƒ if ƒsa, bd … ƒsx, yd for all domain
points (x, y) in an open disk centered at (a, b).
x
y

Local maxima correspond to mountain peaks on the surface z = ƒsx, yd and local minima
correspond to valley bottoms (Figure 14.38). At such points the tangent planes, when they
exist, are horizontal. Local extrema are also called relative extrema.
As with functions of a single variable, the key to identifying the local extrema is a
first derivative test.
FIGURE 14.37 The “roof surface”

A ƒ ƒ x ƒ - ƒ yƒ ƒ - ƒ x ƒ - ƒ y ƒB
Local maxima
1
z = (no greater value of f nearby)
2

viewed from the point (10, 15, 20). The

defining function has a maximum value of
0 and a minimum value of -a on the
square region ƒ x ƒ … a, ƒ y ƒ … a .

Local minimum
(no smaller value
of f nearby)

HISTORICAL BIOGRAPHY FIGURE 14.38 A local maximum is a mountain peak and a local
minimum is a valley low.
Siméon-Denis Poisson
(1781–1840)
THEOREM 10 First Derivative Test for Local Extreme Values
If ƒ(x, y) has a local maximum or minimum value at an interior point (a, b) of its
domain and if the first partial derivatives exist there, then ƒxsa, bd = 0 and
ƒysa, bd = 0.
z
0f
0
0x z  f (x, y) Proof If ƒ has a local extremum at (a, b), then the function gsxd = ƒsx, bd has a local ex-
0f tremum at x = a (Figure 14.39). Therefore, g¿sad = 0 (Chapter 4, Theorem 2). Now
0
0y g¿sad = ƒxsa, bd, so ƒxsa, bd = 0. A similar argument with the function hsyd = ƒsa, yd
g(x)  f (x, b) shows that ƒysa, bd = 0.
0 b h( y)  f (a, y)
a If we substitute the values ƒxsa, bd = 0 and ƒysa, bd = 0 into the equation
y
ƒxsa, bdsx - ad + ƒysa, bds y - bd - sz - ƒsa, bdd = 0
(a, b, 0)
x
for the tangent plane to the surface z = ƒsx, yd at (a, b), the equation reduces to
FIGURE 14.39 If a local maximum of ƒ 0 # sx - ad + 0 # s y - bd - z + ƒsa, bd = 0
occurs at x = a, y = b , then the first
or
partial derivatives ƒxsa, bd and ƒysa, bd are
both zero. z = ƒsa, bd.

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 8/25/04 2:53 PM Page 1029

14.7 Extreme Values and Saddle Points 1029

z Thus, Theorem 10 says that the surface does indeed have a horizontal tangent plane at a lo-
cal extremum, provided there is a tangent plane there.

DEFINITION Critical Point

y An interior point of the domain of a function ƒ(x, y) where both ƒx and ƒy are zero
or where one or both of ƒx and ƒy do not exist is a critical point of ƒ.
x

xy (x 2  y 2 ) Theorem 10 says that the only points where a function ƒ(x, y) can assume extreme val-
z
x2  y2 ues are critical points and boundary points. As with differentiable functions of a single
z variable, not every critical point gives rise to a local extremum. A differentiable function
of a single variable might have a point of inflection. A differentiable function of two vari-
ables might have a saddle point.

y
DEFINITION Saddle Point
x
A differentiable function ƒ(x, y) has a saddle point at a critical point (a, b) if in
every open disk centered at (a, b) there are domain points (x, y) where
ƒsx, yd 7 ƒsa, bd and domain points sx, yd where ƒsx, yd 6 ƒsa, bd. The corre-
sponding point (a, b, ƒ(a, b)) on the surface z = ƒsx, yd is called a saddle point of
the surface (Figure 14.40).

z  y2  y4  x2

EXAMPLE 1 Finding Local Extreme Values

FIGURE 14.40 Saddle points at the
Find the local extreme values of ƒsx, yd = x 2 + y 2.
origin.
Solution The domain of ƒ is the entire plane (so there are no boundary points) and the
partial derivatives ƒx = 2x and ƒy = 2y exist everywhere. Therefore, local extreme values
can occur only where
ƒx = 2x = 0 and ƒy = 2y = 0.
z
The only possibility is the origin, where the value of ƒ is zero. Since ƒ is never negative,
z  x 2  y2
we see that the origin gives a local minimum (Figure 14.41).

EXAMPLE 2 Identifying a Saddle Point

Find the local extreme values (if any) of ƒsx, yd = y 2 - x 2.

x
y
FIGURE 14.41 The graph of the function
ƒsx, yd = x 2 + y 2 is the paraboloid
z = x 2 + y 2. The function has a local
minimum value of 0 at the origin
(Example 1).
Solution The domain of ƒ is the entire plane (so there are no boundary points) and the
partial derivatives ƒx = -2x and ƒy = 2y exist everywhere. Therefore, local extrema can
occur only at the origin (0, 0). Along the positive x-axis, however, ƒ has the value
ƒsx, 0d = -x 2 6 0; along the positive y-axis, ƒ has the value ƒs0, yd = y 2 7 0. There-
fore, every open disk in the xy-plane centered at (0, 0) contains points where the function
is positive and points where it is negative. The function has a saddle point at the origin
(Figure 14.42) instead of a local extreme value. We conclude that the function has no local
extreme values.
That ƒx = ƒy = 0 at an interior point (a, b) of R does not guarantee ƒ has a local ex-
treme value there. If ƒ and its first and second partial derivatives are continuous on R, how-
ever, we may be able to learn more from the following theorem, proved in Section 14.10.

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 9/2/04 11:09 AM Page 1030

1030 Chapter 14: Partial Derivatives

z
y2  x2 THEOREM 11 Second Derivative Test for Local Extreme Values

Suppose that ƒ(x, y) and its first and second partial derivatives are continuous
throughout a disk centered at (a, b) and that ƒxsa, bd = ƒysa, bd = 0 . Then
i. ƒ has a local maximum at (a, b) if ƒxx 6 0 and ƒxx ƒyy - ƒxy2 7 0 at (a, b).
y
ii. ƒ has a local minimum at (a, b) if ƒxx 7 0 and ƒxx ƒyy - ƒxy2 7 0 at (a, b).
x iii. ƒ has a saddle point at (a, b) if ƒxx ƒyy - ƒxy2 6 0 at (a, b).
iv. The test is inconclusive at (a, b) if ƒxx ƒyy - ƒxy2 = 0 at (a, b). In this case,
FIGURE 14.42 The origin is a saddle we must find some other way to determine the behavior of ƒ at (a, b).
point of the function ƒsx, yd = y 2 - x 2.
There are no local extreme values
(Example 2).
The expression ƒxx ƒyy - ƒxy2 is called the discriminant or Hessian of ƒ. It is some-
times easier to remember it in determinant form,

ƒxx ƒyy - ƒxy2 = p p.

ƒxx ƒxy
ƒxy ƒyy
Theorem 11 says that if the discriminant is positive at the point (a, b), then the surface
curves the same way in all directions: downward if ƒxx 6 0, giving rise to a local maxi-
mum, and upward if ƒxx 7 0, giving a local minimum. On the other hand, if the discrimi-
nant is negative at (a, b), then the surface curves up in some directions and down in others,
so we have a saddle point.

EXAMPLE 3 Finding Local Extreme Values

Find the local extreme values of the function
ƒsx, yd = xy - x 2 - y 2 - 2x - 2y + 4.

Solution The function is defined and differentiable for all x and y and its domain has
no boundary points. The function therefore has extreme values only at the points where ƒx
and ƒy are simultaneously zero. This leads to
ƒx = y - 2x - 2 = 0, ƒy = x - 2y - 2 = 0,
or
x = y = -2.
Therefore, the point s -2, -2d is the only point where ƒ may take on an extreme value. To
see if it does so, we calculate
ƒxx = -2, ƒyy = -2, ƒxy = 1.
The discriminant of ƒ at sa, bd = s -2, -2d is
ƒxx ƒyy - ƒxy2 = s -2ds -2d - s1d2 = 4 - 1 = 3.
The combination
ƒxx 6 0 and ƒxx ƒyy - ƒxy2 7 0
tells us that ƒ has a local maximum at s -2, -2d . The value of ƒ at this point is
ƒs -2, -2d = 8.

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 8/25/04 2:53 PM Page 1031

14.7 Extreme Values and Saddle Points 1031

EXAMPLE 4 Searching for Local Extreme Values

z y
Find the local extreme values of ƒsx, yd = xy.

Solution Since ƒ is differentiable everywhere (Figure 14.43), it can assume extreme

values only where
ƒx = y = 0 and ƒy = x = 0.
x
Thus, the origin is the only point where ƒ might have an extreme value. To see what hap-
pens there, we calculate
ƒxx = 0, ƒyy = 0, ƒxy = 1.
The discriminant,
z  xy
ƒxx ƒyy - ƒxy2 = -1,
FIGURE 14.43 The surface z = xy has a
is negative. Therefore, the function has a saddle point at (0, 0). We conclude that
saddle point at the origin (Example 4).
ƒsx, yd = xy has no local extreme values.

Absolute Maxima and Minima on Closed Bounded Regions

We organize the search for the absolute extrema of a continuous function ƒ(x, y) on a
closed and bounded region R into three steps.
1. List the interior points of R where ƒ may have local maxima and minima and evaluate
ƒ at these points. These are the critical points of ƒ.
2. List the boundary points of R where ƒ has local maxima and minima and evaluate ƒ at
these points. We show how to do this shortly.
3. Look through the lists for the maximum and minimum values of ƒ. These will be the
absolute maximum and minimum values of ƒ on R. Since absolute maxima and min-
ima are also local maxima and minima, the absolute maximum and minimum values
of ƒ appear somewhere in the lists made in Steps 1 and 2.
y

B(0, 9)
EXAMPLE 5 Finding Absolute Extrema

y9x
Find the absolute maximum and minimum values of

9 , 9 ƒsx, yd = 2 + 2x + 2y - x 2 - y 2
2 2
x0
on the triangular region in the first quadrant bounded by the lines x = 0, y = 0,
y = 9 - x.
(1, 1)
x
O y0 A(9, 0) Solution Since ƒ is differentiable, the only places where ƒ can assume these values are
points inside the triangle (Figure 14.44) where ƒx = ƒy = 0 and points on the boundary.
FIGURE 14.44 This triangular region is
(a) Interior points. For these we have
the domain of the function in Example 5.

fx = 2 - 2x = 0, fy = 2 - 2y = 0,

yielding the single point sx, yd = s1, 1d. The value of ƒ there is
ƒs1, 1d = 4.

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 8/25/04 2:53 PM Page 1032

1032 Chapter 14: Partial Derivatives

(b) Boundary points. We take the triangle one side at a time:

(i) On the segment OA, y = 0. The function
ƒsx, yd = ƒsx, 0d = 2 + 2x - x 2
may now be regarded as a function of x defined on the closed interval 0 … x … 9. Its
extreme values (we know from Chapter 4) may occur at the endpoints
x = 0 where ƒs0, 0d = 2
x = 9 where ƒs9, 0d = 2 + 18 - 81 = -61
and at the interior points where ƒ¿sx, 0d = 2 - 2x = 0. The only interior point where
ƒ¿sx, 0d = 0 is x = 1, where

ƒsx, 0d = ƒs1, 0d = 3.

(ii) On the segment OB, x = 0 and

ƒsx, yd = ƒs0, yd = 2 + 2y - y 2.

We know from the symmetry of ƒ in x and y and from the analysis we just carried out
that the candidates on this segment are
ƒs0, 0d = 2, ƒs0, 9d = -61, ƒs0, 1d = 3.
(iii) We have already accounted for the values of ƒ at the endpoints of AB, so we need only
look at the interior points of AB. With y = 9 - x, we have
ƒsx, yd = 2 + 2x + 2s9 - xd - x 2 - s9 - xd2 = -61 + 18x - 2x 2.

Setting ƒ¿sx, 9 - xd = 18 - 4x = 0 gives

18 9
x = = .
4 2
At this value of x,

ƒsx, yd = ƒ a , b = - .
9 9 9 9 41
y = 9 - = and
2 2 2 2 2

Summary We list all the candidates: 4, 2, -61, 3, -s41>2d. The maximum is 4, which ƒ
assumes at (1, 1). The minimum is -61, which ƒ assumes at (0, 9) and (9, 0).

Solving extreme value problems with algebraic constraints on the variables usually re-
quires the method of Lagrange multipliers in the next section. But sometimes we can solve
such problems directly, as in the next example.

EXAMPLE 6 Solving a Volume Problem with a Constraint

A delivery company accepts only rectangular boxes the sum of whose length and girth
(perimeter of a cross-section) does not exceed 108 in. Find the dimensions of an accept-
able box of largest volume.

Solution Let x, y, and z represent the length, width, and height of the rectangular box,
respectively. Then the girth is 2y + 2z. We want to maximize the volume V = xyz of the

Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley

4100 AWL/Thomas_ch14p965-1066 8/25/04 2:53 PM Page 1033
Girth  distance
around here
z = 108yz - 2y 2z - 2yz 2
x y
FIGURE 14.45 The box in Example 6.
Copyright © 2005 Pearson Education, Inc., publishing as Pearson Addison-Wesley
14.7 Extreme Values and Saddle Points 1033
box (Figure 14.45) satisfying x + 2y + 2z = 108 (the largest box accepted by the deliv-
ery company). Thus, we can write the volume of the box as a function of two variables.
V = xyz and
Vs y, zd = s108 - 2y - 2zdyz x = 108 - 2y - 2z
Setting the first partial derivatives equal to zero,
Vys y, zd = 108z - 4yz - 2z 2 = s108 - 4y - 2zdz = 0
Vzs y, zd = 108y - 2y 2 - 4yz = s108 - 2y - 4zdy = 0,
gives the critical points (0, 0), (0, 54), (54, 0), and (18, 18). The volume is zero at (0, 0),
(0, 54), (54, 0), which are not maximum values. At the point (18, 18), we apply the Second
Derivative Test (Theorem 11):
Vyy = -4z, Vzz = -4y, Vyz = 108 - 4y - 4z.
Then
Vyy Vzz - V yz2 = 16yz - 16s27 - y - zd2.
Thus,
Vyys18, 18d = -4s18d 6 0
and
C Vyy Vzz - V yz2 D s18,18d = 16s18ds18d - 16s -9d2 7 0
imply that (18, 18) gives a maximum volume. The dimensions of the package are
x = 108 - 2s18d - 2s18d = 36 in., y = 18 in., and z = 18 in. The maximum volume is
V = s36ds18ds18d = 11,664 in.3 , or 6.75 ft3.
Despite the power of Theorem 10, we urge you to remember its limitations. It does not ap-
ply to boundary points of a function’s domain, where it is possible for a function to have
extreme values along with nonzero derivatives. Also, it does not apply to points where ei-
ther ƒx or ƒy fails to exist.
Summary of Max-Min Tests
The extreme values of ƒ(x, y) can occur only at
i. boundary points of the domain of ƒ
ii. critical points (interior points where ƒx = ƒy = 0 or points where ƒx or ƒy
fail to exist).
If the first- and second-order partial derivatives of ƒ are continuous throughout a
disk centered at a point (a, b) and ƒxsa, bd = ƒysa, bd = 0, the nature of ƒ(a, b)
can be tested with the Second Derivative Test:
i. ƒxx 6 0 and ƒxx ƒyy - ƒxy2 7 0 at sa, bd Q local maximum
ii. ƒxx 7 0 and ƒxx ƒyy - ƒxy2 7 0 at sa, bd Q local minimum
iii. ƒxx ƒyy - ƒxy2 6 0 at sa, bd Q saddle point
iv. ƒxx ƒyy - ƒxy2 = 0 at sa, bd Q test is inconclusive.